Privacy Protection (PII)
The Privacy Protection tile lets AI-Public automatically check chat messages for personally identifiable information (PII) before they are sent to the AI. Per category, you configure how the system handles detected data.
What is PII protection?
PII stands for Personally Identifiable Information – data that can directly or indirectly identify a person, such as email addresses, social security numbers, credit card data or phone numbers.
When PII protection is enabled, user chat messages are analysed before reaching the AI. Depending on the configured action per category, data is passed through, anonymised, masked or blocked.
Enabling PII protection
The Privacy Protection tile is located in the management environment.
- Go to the management environment
- Select the Privacy Protection tile
- Enable the Activate PII protection switch
- Configure the desired action per category
- Click Save
While the switch is disabled, chat messages are forwarded unchanged.
Available actions per category
| Action | Description |
|---|---|
| Pass through | The information is forwarded unchanged. Use this for data that poses no risk. |
| Anonymise | The information is fully replaced by a placeholder (e.g. jan@organisation.com → [EMAIL]). The AI does not see the real data. |
| Mask | The information is partially hidden (e.g. ****-****-****-1234). The type is recognisable, but the full value is not. |
| Block | The chat message is completely blocked and not sent to the AI. Use this for highly sensitive information. |
Available categories
| Category | What is detected |
|---|---|
| Email addresses | Addresses such as jan@example.com |
| Credit card numbers | Visa, Mastercard and other cards |
| Social security numbers (BSN) | 9-digit BSN numbers |
| Phone numbers | Dutch and international formats |
| IBAN / Bank account numbers | National and international IBAN numbers |
| IP addresses | IPv4 and IPv6 addresses |
| Passport / ID numbers | Passport and identity card numbers |
| Medical information | Medical terms and health information |
| URLs | Web links and page addresses |
| MAC addresses | MAC addresses of network devices |
Recommended configuration
A common configuration for your environment:
| Category | Recommended action |
|---|---|
| Email addresses | Pass through |
| BSN | Mask |
| Credit card numbers | Mask |
| IBAN | Mask |
| Phone numbers | Mask |
| Medical information | Anonymise |
| IP addresses | Mask |
| URLs | Pass through |
| Passport / ID numbers | Mask |
| MAC addresses | Mask |
How does medical information detection work?
The Medical information category works differently from the other categories. Because medical data is much more varied in nature, AI-Public uses heuristic keyword-based detection.
The system scans chat messages for a fixed list of medical terms in both Dutch and English:
| Dutch | English |
|---|---|
| diagnose / diagnoses | diagnosis |
| medicijn / medicatie | medication |
| allergie / allergieën | allergy / allergies |
| bloeddruk | blood pressure |
| hartslag | heart rate |
| cholesterol | cholesterol |
| diabetes | diabetes |
| insuline | insulin |
| patiënt | patient |
| symptoom / symptomen | symptom |
| behandeling | — |
| operatie | — |
Limitations
- False positives – a sentence such as "The operation of the company went smoothly" may be flagged as medical.
- False negatives – a rare condition not in the list will not be detected.
Use the Anonymise setting for medical information (the default). The exact medical term is replaced by a placeholder, so the AI retains context without seeing the specific data. Use Mask if you only want to partially hide the term.
Read and write permissions
Administrators with read-only access can view the settings but not change them. Contact AI-Public if you want to make changes but do not have sufficient permissions.
See also
- Restrictions – configure usage limits
- System Prompts – guide AI behaviour via system instructions
- Security (privacy-avg-security) – general security overview